运动估计消耗视频编码系统绝大部分的计算量。H.264由于采用多模式运动估计和四分之一精度搜索，不仅增加了整像素运动估计的计算量，而且也大大增加了分像素搜索运算量。随着快速整像素搜索算法的发展，整像素计算量不断降低。因此对于H.264的实时应用，快速分像素运动估计算法变得十分重要。提出了一种全新的基于图像平坦区域预测的分像素运动搜索算法。该算法通过预测图像平坦均匀区域，有效地减少了分像素搜索点数。实验结果表明，与全搜索分像素搜索相比，该算法可以减少22%~63%左右的分像素搜索点数，同时基本不改变图像质量和编码压缩率。

Motion estimation (ME) consumes most of the computational resource in a video encoder system. H.264 adopts multimode ME and quarter pel accuracy fractional pixel search, which brings about large scale of computational cost not only to integer pixel ME but also to fractional pixel ME. However, the computational cost of integer pixel ME has been greatly reduced with the development of fast integer search algorithms. Therefore, fast fractional pixel ME algorithms become significantly important for real time applications. A novel fractional pixel motion estimation based on prediction of flat regions of an image is first proposed in this paper. It reduces fractional pel search points efficiently by predicting flat homogeneous regions of the image. Experimental results show that compared to the full fractional pixel search, the proposed method can reduce 22% to 63% of fractional pixel search points while keeping the image quality and compression rate almost unchanged.